Federal Reserve Governor Christopher J. Waller used his September 29 address at Sibos 2026 in Miami to do something no sitting US central bank official has done at this level of detail: lay out a structural map of what has to change before autonomous software agents can spend money on behalf of businesses and consumers.
His speech, Payments in the Age of AI Agents, broke agentic commerce into two models. Agent-assisted keeps the buyer in control – the AI finds the product, the human pays. Agent-delegated hands the agent authority to shop and pay autonomously within guardrails. The delegated model is where the real commerce velocity shows up, and it is also where the structural problems concentrate.
The Three Walls
Waller named three barriers to scaling agent-delegated payments. First, authentication: the question is no longer whether the buyer is authorized to pay, but whether the agent has been granted authority to pay on the buyer’s behalf. Second, liability: when an agent makes the wrong purchase, who is on the hook? Existing e-commerce frameworks – network rules, consumer protection standards – might adapt, but Waller left room for new approaches. Third, fraud: systems calibrated to human behavioral patterns will need recalibration for the high-frequency, machine-speed patterns of autonomous agents.
Agentic transactions could materially change the frequency and timing of payments, as well as other characteristics.
Waller described trust as “the biggest barrier to scaling agentic commerce, particularly the agent-delegated model.” That framing matters because it matches the structural thesis the payment networks are already building against.
A Market-Structure Problem, Not a Fed Mandate
Waller framed these challenges as a market-structure problem before a payments-infrastructure problem. The industry – not the central bank – is expected to develop the authentication and liability rules. He pointed to Visa Intelligent Commerce and Mastercard Agent Pay as examples of card networks already publishing technical specifications for agent registration and logging cardholder approvals.
The timing tracks. Mastercard expanded Agent Pay on September 30 – the day after Waller’s speech – adding probability scoring for agent-initiated transactions and trust-and-intelligence signals. That is the first network-level attempt to provide what the industry is starting to call a machine-readable record of delegated authority: essentially a digital power of attorney attached to every transaction, identifying who authorized the agent, what it can buy, spending limits, and expiration.
B2B Amplifies the Stakes
While consumer shopping gets the attention, Waller pointed to B2B transactions as potentially better suited for agentic commerce – recurring purchases, approved-supplier rules, and budget limits create natural guardrails for agents. But higher transaction values amplify the financial exposure when an agent makes a mistake or acts outside its authority.
The micropayment layer compounds the complexity. Agents will need to pay for LLM queries, price data feeds, and API calls before completing a purchase. That kind of machine-to-machine transaction volume favors payment rails with lower flat fees – a cost-structure question the industry has not fully resolved.
The Structural Fork
Waller also raised the open-versus-closed question: will agentic commerce operate on interoperable standards that work across e-commerce systems, agent interfaces, and payment methods? Or will merchants require specific AI agents and block others, creating closed ecosystems with more vertical integration? He called the outcome “not clear” but acknowledged it “could significantly influence how the market structure evolves.”
The liability gap the beat has been tracking – from the authentication challenge through the trust infrastructure sprint that saw Mastercard Proto, Visa AI Financial Assistant, and the EPAA working group all launch in the same week – is now named at the highest policy level available. The Fed did not prescribe a standard. It drew the map and pointed at the industry.